Designing a Real-Time Silkworm Cocoon Segregator Using Machine Learning

Silk cocoons are used for various applications and the best ones are mostly available in two shapes – oval and dumbbell. Oval and dumbbell shaped silk cocoons have their own pros and cons. Silk cocoon shipments received are mostly in the mixed form and there is a need to segregate cocoons based on their shapes for best reeling and seeding results. Design of a real-time silkworm cocoon segregator using the concepts of machine learning is proposed in this paper. A single ESP32 camera module is used to capture real-time input with a Neural Network model running on the same module to classify and segregate oval and dumbbell shaped cocoons. Numerous models were designed with their performances evaluated to identify the model with best configuration. The proposed real-time system with best configuration could classify cocoons in 717ms with a maximum recognition accuracy of 99%. The proposed system is an outcome of a funded project with National Silkworm Seed Organization, Central Silk Board under Ministry of Textiles, India.